What Happens when More Starbucks Cafés come to Town?

Page created by Roger Frank
 
CONTINUE READING
What Happens when More Starbucks Cafés come to Town?

 Philip G. Gayle*, Jin Wang**, and Ying Lin***

 This draft: September 26, 2022
 First draft: January 14, 2019

 Abstract

This study first uses a theoretical model to illustrate that the increasing market presence of
Starbucks cafés has two opposing effects on the demand for packaged ground coffee products sold
via grocery stores, i.e., “demand-increasing” and “business-stealing” effects, the relative strengths
of which determine the ultimate market outcomes. We then empirically examine how these effects
influence the equilibrium market outcomes. The empirical findings reveal that market presence of
an additional Starbucks café increases the prices and quantities sold of Starbucks and non-
Starbucks packaged ground coffee products in grocery stores and suggest a relatively dominant
“demand-increasing” effect. Furthermore, we find evidence of net welfare gains associated with
Starbucks café entry in local markets.

Keywords: Starbucks Cafés, Grocery Stores, Retail Packaged Ground Coffee, Business-stealing
 Effect, Demand-increasing Effect

JEL Classification Codes: L13, D12, L66, M30

This paper benefited greatly from discussions and comments from conference participants at the 97 th annual
conference of Western Economic Association International (WEAI), Portland, OR, June 29 – July 3, 2022; participants
at the 91st annual meeting of the Southern Economic Association (SEA), Houston, TX, November 19 – 22, 2021; and
participants at the 19th annual International Industrial Organization Conference (IIOC), April 30 – May 2, 2021. Any
remaining errors are our own.
*
 Kansas State University, Department of Economics, 322 Waters Hall, Manhattan, KS 66506; Voice: (785) 532-4581;
 Fax: (785) 532-6919 email: gaylep@ksu.edu; Corresponding author.
**
 Uber Technologies, Inc., 1455 Market St. Ste 400, San Francisco, CA 94106; email: jinwangccer@gmail.edu.
***
 Belhaven University, School of Business, Box 310, Jackson, MS 39202; Voice: 601-968-8982; Email:
 ylin@belhaven.edu.
1. Introduction

 The increasing number of consumer goods and services in recent years is a consequence of
the diligent efforts that firms have been putting on proliferating their product lines to meet
consumers’ ever-changing needs and preferences. When a firm extends its product line, it needs
not only to decide which type of new products to launch, but also to carefully coordinate the
relationship between the new products with the existing products. On the one hand, if two products
are positioned too “closely” with each other in the product line, for example, having similar quality
or functionality, a business-stealing effect can easily occur. This may shift the existing consumers
between products resulting in gains for one product and losses for the other. On the other hand, a
multi-product firm wants to leverage the strong favorable reputation of established products and
brands to promote the sales of new products. This idea is well expressed by Michael Conway, a
Starbucks Corp. executive vice president, when he explained the company’s strategy of launching
packaged ground coffee in grocery stores: “Because of the impression we make every day in the
cafés, we don’t have to work as hard when we launch new products”.1
 The coexistence of business cannibalization/stealing and demand-increasing effects
complicates the interrelationship between products as well as the business decisions of a firm
because a strategy which is meant to be applied to one product category can have an ambiguous
impact on the profit of another product category of the firm, and sometimes the impact can even
spill over to related products of competing firms. This calls for firms and researchers to consider
these two types of effects across products and across firms when evaluating strategies for changes
in product lines. Our empirical focus in this study is to analyze the interrelationships between
Starbucks café outlet line and its retail packaged ground coffee products line in the grocery chains.
 Starbucks, the Seattle-based coffee chain, since it opened its first coffee store in 1971, has
aggressively expanded its café brand by locating thousands of café shops across the world. In
contrast to the rapid growth of café shops, it took Starbucks a long time to launch and grow its
business of packaged coffee products retailed in grocery stores. In 1989, 18 years after it was
founded, the company began providing private-label coffee in Costco, a wholesale grocery chain,
under the Kirkland Signature Brand. In 1998, Starbucks partnered with Kraft, a food

1
 “Starbucks' grocery gambit” by Beth Kowitt, Fortune, December 5, 2013. http://fortune.com/2013/12/05/starbucks-
grocery-gambit/

 1
manufacturing firm with favorable brand recognition in grocery stores, to distribute its packaged
coffee in grocery stores, but this partnership was terminated by Starbucks in late 2010 as Starbucks
alleged that Kraft failed to market its brands in grocery stores satisfactorily. 2 Subsequently,
Starbucks decided to own and control its packaged ground coffee products by itself. “Shouldn’t
our shareholders benefit from our ability to build a brand inside Starbucks?” the company’s CEO
Howard Schultz said, “as opposed to selling a category or product that is something that we don’t
own or have equity in?” 1
 To explore the consumer brand awareness accumulated by its café shops, Starbucks placed
so-called “Signature Aisles” in grocery stores to “feature bags of coffee under a sign with
Starbucks’ distinctive mermaid logo”.2 Moreover, Starbucks extended its loyalty program from
café shops to grocery store aisles. In particular, the loyalty program members who buy Starbucks
ground coffee products in grocery stores can earn rewards points that can be redeemed for coffee,
food and merchandise in Starbucks cafés. Those who redeem rewards points in Starbucks cafés
can receive coupons for their future purchases of Starbucks ground coffee products in grocery
stores. The fact that Starbucks diligently strives to leverage the strong reputation and value created
by its café shops when promoting its packaged ground coffee products in grocery stores makes it
interesting to investigate the interactions of these two streams of product lines.
 The key objective of this study is to examine the market effects associated with the
increasing presence of Starbucks café shops on the U.S. retail packaged ground coffee products in
the grocery channel. Specifically, the study addresses the following questions: (i) How does
Starbucks business strategy of establishing café shops in local markets influences the prices and
quantities of its own and its competitors’ retail packaged ground coffee products sold in grocery
stores located in these markets? (ii) Considering the countervailing impacts of the “demand-
increasing” and “business-stealing” effects on retail packaged coffee products induced by the
market entry of a new Starbucks café shop, which of the two effects dominate in determining
equilibrium market outcomes? and (iii) What is the welfare effect associated with entry of a
Starbucks café shop in a local market?
 The analyses begin by using a theoretical model to illustrate the “business-stealing” and
“demand-increasing” effects of Starbucks café shops on consumers’ quantity demand for

2
 “Starbucks Aims For Grocery Store Supremacy With New Signature Aisle” by Rachel Tepper, Huffton Post, April
26, 2013. https://www.huffingtonpost.com/2013/04/26/starbucks-grocery-store-aisle_n_3157075.html

 2
packaged ground coffee of Starbucks and its competitors. Specifically, on a purchase occasion, a
potential consumer has the option to either purchase packaged ground coffee from grocery stores
and make a drink by themselves, visit a Starbucks café shop to purchase and enjoy barista-prepared
coffee, or choose neither of these options. The model assumes the proportion of grocery coffee
shoppers decreases with the number of Starbucks café shops: when there are more Starbucks café
shops in a market, consumers more easily patronize a café to grab a cup of barista-brewed coffee
and therefore become less likely to buy packaged coffee from supermarkets. This is referred to as
a “business-stealing” effect.
 Meanwhile, for those consumers who continue to buy ground coffee in grocery stores, the
increasing presence of Starbucks café shops also shifts their relative preference for Starbucks and
competing brands, e.g., Folgers, of packaged ground coffee in two distinct ways. First, Starbucks
café shops offer consumers opportunities to try a variety of coffee drinks and explore their
preferences over different types or flavors of coffee. Consumers then can use the packaged ground
coffee purchased from grocery stores to make their ideal coffee drink. This “demand-increasing”
effect may boost consumers’ demand for packaged ground coffee products of both Starbucks and
Folgers if the two brands do not fully cover the market of grocery coffee shoppers. Second, the
presence of Starbucks café shops may increase consumers’ perceived utility of Starbucks packaged
ground coffee in a magnitude higher than that of Folgers packaged ground coffee, driven by
increased consumer awareness of the Starbucks brand as well as the benefits associated with the
Starbucks loyalty program. Such a “demand-increasing” effect may enhance the demand for
Starbucks branded products disproportionately more than Folgers branded products.
 Depending on the relative magnitudes of the above two opposing effects, a new Starbucks
café shop available in the market likely has either a net positive or a net negative effect on the
demand of retail packaged coffee products, Starbucks brand and competing non-Starbucks brands,
in the grocery channel. Our theoretical analysis shows that due to countervailing forces of the
“business-stealing” and “demand-increasing” effects, an increase in the number of available
Starbucks café shops in the market has an ambiguous impact on the equilibrium quantities sold of
retail packaged ground coffee for both Starbucks brand products and competing non-Starbucks
brand products. The theoretical ambiguity, therefore, makes it necessary to examine the impacts
in real-world settings. A similar countervailing tension between “demand-increasing” versus
“business-stealing” effects is empirically examined by Berry and Waldfogel (1999) in the context

 3
of the provision of various public radio programming. Accordingly, we specify a structural
econometric model to empirically examine the market demand and supply of Starbucks and non-
Starbucks packaged ground coffee products sold in grocery stores. The estimated model is
subsequently used to simulate new market equilibrium outcomes based on an assumed
counterfactual change in the number of Starbucks café shops present in the relevant market.
 For the empirical analyses we use a standard discrete-choice demand model, and assume
that competing firms set prices according to a static Bertrand-Nash price-setting game. The
demand model is specified to allow the number of Starbucks café shops in the relevant market to
influence consumers’ mean utility obtained from consuming package ground coffee products sold
in grocery stores, as well as influence consumers’ mean utility obtained from choosing the outside
option. The outside option of our demand model includes consumers’ choosing to consume
freshly-brewed coffee from a local café shop. The estimated demand model and supply-side
equations implied by the assumed price-setting behavior of firms are used jointly to simulate the
new market equilibrium that would result from the counterfactual entry of an additional Starbucks
café shop in the relevant market. A comparison of actual market outcomes with model-predicted
market outcomes resulting from the counterfactual entry of an additional Starbucks café shop
reveals the extent to which an additional Starbucks café shop influences prices and quantities sold
of Starbucks’ and non-Starbucks brand packaged ground coffee products sold in local grocery
stores, as well as the impact on consumers’ welfare.
 The demand estimates suggest that each additional Starbucks café shop in the local market
increases consumer demand for both Starbucks and competing non-Starbucks packaged ground
coffee products in the grocery chains. The counterfactual experiments reveal that the entry of an
additional Starbucks café shop increases both prices and quantities sold of Starbucks and
competing non-Starbucks packaged ground coffee products. The findings provide empirical
evidence that the increasing market presence of Starbucks cafés exhibits a “demand-increasing”
effect that dominates the “business-stealing” effect on the sales of retail packaged coffee products,
which ultimately expand the retail packaged coffee market segment. Furthermore, the size of the
market-expansionary effect attenuates as the market becomes increasingly saturated with
Starbucks café shops. Last, based on the predicted changes in money-metric consumer surplus, we
find net welfare gains associated with market entry of a new Starbucks café.

 4
The remainder of the paper proceeds as follows. In the next section, we briefly review the
relevant literature. In section 3 we discuss key empirically testable insights from a simple
theoretical model that we specify and analyze. Section 4 discusses the data used in the empirical
analysis. Section 5 describes the empirical model, as well as estimation and identification of
parameters in the empirical model. Section 6 presents and discusses the empirical results. Section
7 concludes the paper.

2. Related Literature

 This paper joins the literature of “umbrella branding” and “brand extension”. Brand is one
of the most important assets of firms. It is very common for a multi-product firm to label several
products under the same brand name, a practice which is termed “umbrella branding” or “brand
extension” [Aaker and Keller (1990), Pepall and Richards (2002)]. Umbrella branding is a
justifiable firm strategy posited by two distinct theories in the literature. First, it has been shown
that umbrella branding is a rational strategy within the theoretical framework of adverse selection
models. Specifically, using an adverse selection model it can be shown that firms, under some
conditions, will find it optimal to leverage the reputations created by existing products and brands
to signal the high quality of new products [Wernerfelt (1988); Choi (1998); Cabral (2000); Miklos-
Thal (2012); Moorthy (2012)]. Second, it has also been shown that umbrella branding is a rational
strategy within the theoretical framework of moral hazard models. In particular, it has been
assumed within a moral hazard model framework that firms’ product quality choice is endogenous,
and scholars use this theoretical framework to demonstrate that umbrella branding may lead to a
larger scope of high-quality investment when a firm simultaneously chooses the qualities of
multiple products [Andersson (2002); Cai and Obara (2006); Hakenes and Peitz (2008); Cabral
(2009); Rasmusen (2016)].
 There also exists a rich body of empirical work which documents the spillover of brand
image among products under the same brand name [see Keller and Lehmann (2006) for an
overview]. For instance, Sullivan (1990) finds the Audi 5000's alleged sudden-acceleration defect
has a significant negative impact on the demand for Audi 4000 and Quattro, whereas the launching
of Jaguar’s new model leads to an increase in demand for its old model as a result of advertising
used to promote the new model. Erdem (1998) proposes a model in which consumers’ quality
perceptions regarding a brand in one product category are affected by their experiences with the

 5
same brand in another category. The model is estimated on panel data for toothpaste and
toothbrushes and the results show that the correlation coefficient of consumers’ prior beliefs about
the qualities of two umbrella products is 0.882, which is taken as an empirical support to the
signalling theory of umbrella branding.
 Our paper investigates spillover effects of Starbucks café shops on the demand and pricing
of its umbrella branding products, Starbucks and its competing brands packaged ground coffee
products sold via the grocery chains. This research is different from previous empirical work of
umbrella branding. Specifically, we allow the co-existence of both positive and negative spillover
effects associated with a firm’s action in a given product category. A positive spillover effect
increases the demand of related products, while a negative spillover effect decreases the demand
of related products. Furthermore, we allow the relative magnitudes of these opposing effects to
vary by the number of Starbucks café shops in a given market. Before performing empirical
analyses, we propose a theoretical framework that demonstrates how the number of Starbucks café
shops may influence equilibrium prices and quantities of Starbucks and non-Starbucks packaged
ground coffee products. This theoretical analysis sheds light on the empirical results we later find.
 Spillover effect has also received a lot of attention in the advertising literature. Our paper,
thus, is also related to that strand of literature. For instance, Garthwaite (2014) studies the
economic effect of celebrity endorsements in the publishing sector. The study finds that celebrity
endorsements increase consumers’ purchases of endorsed books and generate spillover benefits
for the non-endorsed titles written by an endorsed author. However, the aggregate adult fiction
sales fall with endorsements, which suggests endorsements are more of a “business-stealing” effect
type of advertising in this case. Chae et. al. (2016) investigate the spillover effects of seeded
marketing campaigns (SMCs) on the generation of “word of mouth” (WOM) at the brand and
category levels. Using data on cosmetics brands, they find brand- and category-level WOM
spillover effects. We focus on another important dimension of firms’ strategies, the location
choices of company-branded retail stores, and its spillover effects on the “umbrella branding”
products sold through grocery channels.
 There is a rich literature in empirical industrial organization in which researchers explicitly
model the market entry decision/strategy of firms. Models in this literature are typically built on
the assumption that a firm will optimally choose to enter a market if its expected variable profits
are sufficient to cover its sunk market entry cost. Seminal papers in this literature include Berry

 6
(1992) and Bresnahan and Reiss (1990, 1991). Even though in this paper we examine the impact
that market entry of a Starbucks café has on packaged ground coffee products sold in grocery
stores, it is beyond the scope of the analysis to explicitly model the market entry decision/strategy
of Starbucks cafés. Accordingly, unlike the local market competition analysis associated with the
market entry of Wal-Mart stores studied in Jia (2008), our study does not measure sunk entry cost
in comparison to the variable profits associated with establishing a Starbucks café, nor do we
attempt to measure the potential net benefits (chain effects) to Starbucks’ café line of business
associated with establishing multiple cafés across neighboring markets.

3. Theoretical Framework and Insights

 Suppose there are two coffee manufacturing firms that correspond to two brands, Folgers
(F) and Starbucks (S), respectively, each of which produces a packaged ground coffee product and
sells through grocery stores. Starbucks also serves consumers fresh-made coffee beverage through
its company-operated or franchised café shops. On a given purchase occasion, a potential
consumer has the option to either purchase packaged ground coffee from a grocery store and make
a drink themself, visit a Starbucks café shop to enjoy barista-prepared coffee, or choose neither of
these options. We normalize the number of potential consumers in a market to a measure of 1, with
a proportion ∈ (0,1) of them choosing to purchase packaged ground coffee from a grocery
store on a given purchase occasion. Therefore, 1 − of the consumers choose not to purchase
packaged ground coffee from a grocery store on the given purchase occasion. The proportion of
consumers choosing not to purchase packaged ground coffee from a grocery store on the given
purchase occasion may have chosen to visit a Starbucks café shop to purchase barista-prepared
coffee, visit some other café shop to purchase barista-prepared coffee, or not purchase coffee of
any type.
 Business-stealing effect: First, we capture the “business stealing” effect by assuming the
number of consumers that buy ground coffee products from grocery stores is determined by the
following function:
 ( ) (1)
where (∙) is a decreasing function of the number of Starbucks café shops , i.e., ′ ( ) < 0.
The idea is that when additional Starbucks café shops enter a market, consumers can more easily
stop by one café to grab a cup of coffee and thus reduce the purchases of packaged ground coffee

 7
from grocery stores. This “business-stealing” effect likely decreases demand for both Starbucks
and Folgers packaged coffee products in grocery stores.
 Demand-increasing effect: We then model the “demand-increasing” effect by allowing
the number of Starbucks café shops to shift a consumer’s utility of purchasing the two brands of
packaged ground coffee in grocery stores:
 = 0 + (1 + ) ( ) − − (2)
 = 0 + ( ) − (1 − ) − (3)
where 0 denotes the intrinsic value of ground coffee. (1 + ) ( ) and ( ) are the
incremental value a consumer attaches to the packaged ground coffee of Starbucks and Folger
respectively, both of which increase with the number of Starbucks café shops, , but at different
rates: (1 + ) ′ ( ) > ′ ( ) > 0. The parameter ≥ 0 measures the extent to which Starbucks
become more favorable for consumers than Folgers as increases. The motivating idea is that
the presence of Starbucks café shops may enhance consumers’ valuations for packaged coffee
products. A reason is that Starbucks café shops offer coffee drinkers opportunities to try a variety
of coffee drinks and explore their preferences over different types/flavors of coffee. Coffee
drinkers can then use the packaged coffee purchased from the grocery stores to make their ideal
coffee drinks.
 We assume that consumers have heterogeneous preferences with respect to the products of
Starbucks and Folgers. Specifically, Starbucks and Folgers are respectively located on the
endpoints 0 and 1 of a Hotelling line of length one. The consumers who may purchase packaged
ground coffee from grocery stores are uniformly distributed along the line and each has a location
indexed by ~ [0,1]: the smaller is, the more the consumer prefers Starbucks’ products. The
unit “transportation cost” is denoted as ∈{ , } , which takes two values > > 0, depending
on how consumers perceive the two brands. Type- consumers are loyal consumers who strongly
prefer one brand and are reluctant to switch to the other brand, while type- consumers are
“switchers” who have a lower “transportation cost” to reach both brands. We assume type- 
consumers take a proportion of among all the ( ) consumers who buy packaged ground
coffee. To simplify the analysis, we also assume that each brand charges a uniform price to
different types of consumers, and denote the price of Starbucks as and the price of Folgers as
 , respectively.

 8
The specification in equation (2) and equation (3) above capture the idea that the “demand-
increasing” effect benefits both the sales of Starbucks and Folgers retail packaged coffee. This
effect is likely to be more pronounced for Starbucks products than Folgers’ for at least two reasons.
First, the presence of Starbucks cafés increases consumers’ awareness of Starbucks brand name,
incentivizing them to choose Starbucks’ packaged ground coffee from among the packaged coffee
products offered in grocery stores. Second, Starbucks loyalty program rewards its member
consumers who purchase Starbucks products from groceries bonus points, which can be redeemed
for free coffee/food/merchandise in local Starbucks café shops. Consumers who redeem rewards
points can further get coupons for their future purchases of Starbucks packaged coffee. A larger
number of Starbucks café shops facilitates consumers to redeem Starbucks rewards points and
increases the benefits associated with the purchases of Starbucks packaged coffee.
 The technical details of the equilibrium analysis and comparative statics generated from
the theoretical framework are described in the Appendix. The comparative statics analysis shows
that as the number of Starbucks café shops increases, the coexistence of the two countervailing
effects on consumer demand leads to an indeterminate overall change in the equilibrium quantity
sold for Starbucks packaged coffee products. The sign of the total effect depends on the relative
dominance between the negative “business-stealing” effect and the positive “demand-increasing”
effect.
 As for the competing brand, Folgers, our theory predicts that an increasing number of
Starbucks café shops also has a mixed effect on its equilibrium quantity sold. The sign of the
overall demand impact depends on the relative sizes of the two consumer segments. The “business-
stealing” effect negatively affects Folger’s sales for sure, while the “demand-increasing” effect
can either positively or negatively affect sales depending on whether the proportion of type- 
consumers exceed a certain threshold, ∗ .
 On the one hand, when the proportion of type- consumers exceeds this threshold (i.e., the
local market is relatively dominated by type- consumers), the “demand-increasing” effect is
positive, weakening the negative “business-stealing” demand impact. This is intuitive, as in the
market segment for type- consumers, Starbucks and Folgers each act as local monopolist to their
respective loyal consumers. The “demand-increasing” effect stimulates the demand of the loyal
consumers for each brand. Therefore, the overall effect on the competing brand of an increase in

 9
the number of Starbucks café shops in markets dominated with type- consumers will depend on
the relative strength of the above two countervailing effects.
 On the other hand, when the proportion of type- consumers is instead smaller than the
threshold (i.e., market is relatively dominated by type- consumers), the “demand-increasing”
effect is negative, reinforcing instead of countering the negative “business-stealing” demand
impact. For type- consumers, the increasing presence of Starbucks cafés, although enhances these
consumers’ utility for each brand, has a larger positive impact on Starbucks packaged coffee. This
in fact disadvantages Folgers during its competition with Starbucks for type- consumers, some
of whom may switch to consume Starbucks packaged ground coffee, resulting in a negative
“demand-increasing” effect. Therefore, the overall effect on the competing brand demand of an
increase in the number of Starbucks café shops in markets dominated with type- consumers is
negative.
 Given the ambiguous effect on equilibrium quantities of packaged ground coffee products
sold in grocery stores induced by the increased market presence of Starbucks café shops, the
theoretical results call for an empirical study which investigates the signs of the impacts in real-
world markets. We undertake such an empirical study throughout subsequent sections of the paper.

4. Data and descriptive analysis

4.1 The data

 We focus our empirical analysis on U.S. retail packaged ground coffee sold in grocery
stores during the sample periods from 2008 through 2012. 3 According to the store opening
information reported in the annual reports provided by Starbucks Corp., we calculate the annual
growth rates of Starbucks café presence in U.S. domestic and international markets, respectively,
and plot these growth rates in Figure 1. The plots reveal that Starbucks café shop count slowly
picked up in 2010, from the massive café shop closures between 2008 and 2009 (The New York
Times4). Specifically, both the domestic and global growth rates of Starbucks cafés experienced a
vast rebound of 3.16% in the U.S. and 6.25% worldwide from 2011 to 2012; and since then,

3
 Retailers of the packaged ground coffee products in our main dataset include supermarkets and drug stores. We
generically refer to these retailers as “grocery stores” throughout the paper.
4
 Starbucks Corp. announced more than 600 store closures and a significant financial loss resulted from the global
economic crisis of 2007-2008. See the relevant article which was retrieved from the New York Times in the following
URL:https://www.nytimes.com/2008/07/02/business/02sbux.html

 10
Starbucks Corp. continues to sustain a moderate growth in terms of café shop expansions
nationwide.

 Figure 1: Growth Rates of Starbucks Café Shops in the U.S. and worldwide

 Source: Authors’ calculation based on Starbucks annual reports.

 We obtain the location data of Starbucks cafés in the U.S. during the sample periods from
AggData, a marketing firm which collects business locational data. The dataset contain
information about the complete café addresses including zip codes and states, business hours, and
in-store wireless status. We define a market as the combination of period (year-month) and grocery
store where various brand packaged ground coffee products are sold. Using the location data, we
create a variable, Nshop, that measures the extent of Starbucks café presence in the local market.
Nshop is the number of Starbucks cafés (either corporate-owned or franchised) located within the
zip code area where the relevant grocery store is located.
 Information about packaged ground coffee products sold in grocery stores is sourced from
the Information Resources Inc. (IRI) academic database [Bronnenberg et al. (2008)], a weekly
scanner dataset available for years 2008 through 2012. The working sample contains consumer
purchases of ground coffee products that are packed in bulk in bags or canisters. Detailed
information for each weekly observation includes: weekly unit sales (in ounces); revenue from

 11
these unit sales; and various attributes that are used to delineate a product in this analysis, such as
brand name, 5 package net weight (in ounces), organic feature, caffeine level, and packaging
material. Details of the sample construction and various product attribute definitions are presented
in Appendix B.
 A product within a defined market is considered as the unique combination of the various
product attributes listed above. The weekly observations are then aggregated to monthly frequency
according to the product and market definition. The “quantity” variable for a defined product is
the sum of weekly unit sales in a month; and the “price” variable for a defined product is the mean
of the weekly average unit price which is obtained by dividing the revenue from weekly unit sales
by the weekly unit sales. The monthly aggregation reduces the data sample to have 402,330 records.
 Our computation of potential market size for each defined market is inspired by the
“potential market factor” method in Ivaldi and Verboven (2005). According to the 2016 coffee
consumption survey provided by the National Coffee Association (NCA), 56% to 64% of the
surveyed population (with an average of 59% over the sample periods) reported they consume
home-brew coffee daily. We assume the retail packaged coffee quantity sales in the data reflect
59% of total quantity that could be potentially purchased by the entire population in a defined
market. Consequently, a market’s potential size is simply the total quantity sales across all products
in the market multiplied by the “potential market factor”. The “potential market factor” in our case
is the reciprocal of 59%. As such, the observed product shares are obtained by dividing the relevant
product’s quantity by this measure of potential market size.
 Summary statistics of the data are reported in Table 1. The retail packaged ground coffee
products in the data sample are sold across 113 grocery stores located in 96 zip code areas. The
average price is about 50 cents per ounce, and 324 ounces of a typical product are sold per month.
The summary statistic for the zero-one dummy variable, Starbucks, implies that about 10% of the
packaged ground coffee products in the sample belong to the Starbucks brand.

5
 Among all the coffee brands in the subsample, Folgers (23.6% of total dollar sales), Starbucks (15.3%), Private Label
(12%), Maxwell House (7%), and Dunkin Donuts (6.3%) are the top five brands with the largest dollar sales over the
sample periods.

 12
Table 1: Summary Statistics
 Std.
Variable Mean Min Max
 Dev.
Price ($/ounce) 0.522 0.242 0.029 2.31
Product Quantity Sold (ounces) 323.847 658.652 7 97,803.8
Starbucks dummy (1 if products belong to Starbucks brand) 0.105 0.307 0 1
Nshop (number of Starbucks café shops in a local market) 4.313 3.436 0 22
Caffeine (grams per ounce of dry coffee grounds) 1.468 0.724 0 2.232
Package Weight (total ounces of dry coffee grounds in a
 17.777 9.755 6 80
package)
Organic dummy (1 if products are organic) 0.058 0.234 0 1
Product market shares (all inside goods) 0.01 0.017 0.000023 0.43
No. of zip code areas 96
No. of grocery stores 113
No. of manufacturers 130
No. of brands 165
No. of defined markets 6,780
No. of observations 402,330
* Prices are adjusted to 2012 dollars.

 Nshop is the count of Starbucks café shops in a defined market. Nshop in our data varies
both over time in a given local market and across markets at a given time. These time series and
cross-sectional variations identify the impact of Nshop on demand for the retail packaged coffee
products in grocery stores. The larger is the value of Nshop for a given market, the greater the
penetration of Starbucks café shops in the local retail packaged coffee market. The data summary
shows that an average of 4 Starbucks café shops locate within the zip code neighborhood of a
typical grocery store where packaged ground coffee products are sold.
 Caffeine measures the caffeine content in grams per ounce of dry coffee grounds. Package
Weight measures the net weight in ounces of coffee grounds in the package, a variable that captures
the impact on demand of consumers’ product package size preferences. Organic is a zero-one
dummy variable capturing heterogeneity in consumers’ preference for organic versus non-organic
coffee products.

4.2 Descriptive analysis

 Before turning to the structural empirical model, we first present some descriptive evidence
regarding the demand impacts on retail packaged coffee product sales associated with the market
presence of Starbucks café shops. In Table 2, we show the regression results that describe
correlations between retail packaged ground coffee product quantities sold and various factors
influencing the quantities sold. In these regressions, we control for local demographics, including

 13
the mean home values, median personal income for adult population (male and female), adult
population at the county-level, adult population at the zip code-level, the fraction of the zip code
population aged 15-64, and the fraction older than 64 years.6
 The positive and statistically significant coefficient estimate on Nshop reveals that
quantities sold of non-Starbucks packaged coffee products in grocery stores increase with the
number of local Starbucks café shops. Furthermore, the number of Starbucks café shops has a
greater positive impact on quantities sold of Starbucks packaged coffee products relative to non-
Starbucks packaged coffee products in local grocery stores, as evidenced by the positive and
statistically significant coefficient estimates on variables Nshop and Starbucks × Nshop,
respectively.
 The positive and statistically significant coefficient estimates on variables Caffeine and
Packaged Weight suggest that sales of packaged coffee increase with caffeine content and package
size, respectively. However, the negative and statistically significant coefficient estimate on the
Organic dummy variable reveals that organic coffee products have lower sales relative to non-
organic coffee products. The negative sign of the coefficient estimate on the competition variable
measured by the number of competing products offered in the market comply with economic
intuition. Specifically, the quantity sold of an individual product tends to decline on average as the
market becomes more competitive.
 Results from the “reduced-form” regressions in Table 2 provide some preliminary evidence
on the potential impacts of the market presence of additional Starbucks café shops on unit sales of
Starbucks and competing non-Starbucks brands of packaged coffee sold in grocery stores.
However, the “reduced-form” regressions in Table 2 are not able to disentangle key driving forces,
“demand-increasing” versus “business-stealing” effects, of the net impact on unit sales of
packaged ground coffee sold in grocery stores caused by the increasing market presence of
Starbucks café shops. Accordingly, we now turn to specifying a structural empirical demand model
designed to separately identify these key driving forces.

6
 Mean home values are measured by zip code-level typical home values obtained from Zillows.com research data
page. We use the mean home values to approximate market average wealth level. Guler (2018) finds a significant
positive effect of consumer wealth level on the number of visits to Starbucks cafés. County-level population and
median income as well as zip code-level population are obtained from the U.S. Census Bureau database.

 14
Table 2: Descriptive Regressions
 Dependent Variable: Log (Quantity Sold)
 Variable (1) (2) (3) (4)
 Nshop 0.0066*** 0.0057*** 0.0065*** 0.0057***
 (0.0017) (0.0017) (0.0017) (0.0017)
 Starbucks*Nshop 0.038*** 0.0379*** 0.038*** 0.038***
 (0.0017) (0.0017) (0.0017) (0.0017)
 Caffeine 0.399*** 0.399*** 0.399*** 0.399***
 (0.0024) (0.0024) (0.0024) (0.0024)
 Package Weight 0.0488*** 0.0488*** 0.0488*** 0.0488***
 (2.09E-04) (2.09E-04) (2.09E-04) (2.09E-04)
 Organic -1.057*** -1.057*** -1.057*** -1.057***
 (0.0087) (0.0087) (0.0087) (0.0087)
 Number of competing products -0.0038*** -0.0036*** -0.0037*** -0.0036***
 (2.10E-04) (2.12E-04) (2.10E-04) (2.12E-04)
 Mean home value 0.0034 0.0053 0.0048 0.0053
 (0.0055) (0.0055) (0.0055) (0.0055)
 County population age>=16 0.0104
 (0.0074)
 Median income for age>=16 0.0331 0.165
 (0.273) (0.275)
 Fraction county male population age>=16 -1.253** -1.396** -1.154**
 (0.564) (0.568) (0.567)
 Median income for male age>=16 -0.721*** -0.723***
 (0.217) (0.217)
 Median income for female age>=16 0.829*** 0.828***
 (0.287) (0.288)
 Fraction zip code population age 15 to 64 0.0057 0.0063*
 (0.0037) (0.0037)
 Fraction zip code population age>=65 0.0094* 0.0091*
 (0.0051) (0.0051)
 Constant 3.649*** 4.564*** 4.009*** 3.948***
 (0.173) (0.330) (0.474) (0.477)
 Number of Observations 402,330 402,330 402,330 402,330
 R-squared 0.403 0.403 0.403 0.403
 F 904.3 901.4 898.3 895.4
 Notes: ***p
Second, we use the estimated structural demand parameters along with an assumed oligopolistic
model of supply to simulate new market equilibrium outcomes resulting from the counterfactual
market presence of an additional Starbucks café shop. A comparison of simulated market
equilibrium outcomes with actual outcomes in the data reveals how equilibrium prices, consumer
demand and welfare are predicted to change with the market presence of an additional Starbucks
café shop.

5.1 Demand

 In each market and period , consumer is assumed to face + 1 product purchase
alternatives indexed by = 0, 1, … , , where alternatives = 1, … , are the various packaged
ground coffee products available for purchase in grocery stores located in market during period
 , i.e., the “inside” goods of the demand model [See Gayle and Lin (2022) for similar modelling
of retail packaged ground coffee demand], while = 0 represents consumers’ “outside”
option/good in market during period .7 The indirect utility consumer obtains from choosing
product is a function of observed and unobserved (to the researchers) non-price product attributes,
price, and individual consumer characteristics that influence individual preferences. We follow a
similar specification of the conditional indirect utility function in Nevo (2000a, b):
 = + + ,1 ℎ + ,2 ℎ × + + (4)
where is a vector of observable product attributes faced by all consumers in the market and
 the associated vector of individual-specific marginal utilities of respective product attributes in
 ; is the price of product assumed to be the same for all consumers in market at time
 ; and is the individual-specific marginal disutility of price.
 Similar to equation (2) and equation (3) in our theoretical framework, our structural
empirical model framework specifies that consumer ′ conditional indirect utility in equation (4)
is a function of the number of Starbucks café shops, ℎ , present in the relevant market with
associated parameters ,1 and ,2 . Consistent with our discussions above in describing the
theoretical framework, parameters ,1 and ,2 in our empirical specification enable measuring
the sign and magnitude of the “demand-increasing” effect of Starbucks café shop presence on non-

7
 The outside option is a composite of several alternatives such as buying other coffee substitutes (e.g. instant, whole
bean, ready-to-drink coffee beverages, etc, which are sold in the grocery stores), buying freshly-brewed coffee from
a local coffee shop, or simply not consuming coffee.

 16
Starbucks brands and Starbucks brand of packaged ground coffee sold in grocery stores.
Specifically, ,1 and ,2 measure the marginal utility impact of the presence of Starbucks café
shops on demand for non-Starbucks brands and Starbucks brand of packaged ground coffee sold
in grocery stores, respectively.
 Similar to Nevo (2000a, b), we model = + ℎ + + + Δ as a
composite of product attributes that are observable to consumers and firms, but unobservable to
the researchers, where , ℎ , and are year, month, grocery store, and brand fixed
effects, respectively; and Δ is left to be the econometric error term. Last, is a mean-zero
idiosyncratic error term that is assumed to follow an independent and identically distributed
extreme value type I density.
 The demand system is completed with the specification of an outside option/good, which
in the context of our demand model includes situations in which the consumer chooses to purchase
freshly-brewed coffee from a local café shop instead of packaged ground coffee from the grocery
store. The indirect utility for consumer selecting the outside option is:
 0 = 0 + 0 + 0 (5)
where the mean utility from the outside option, 0 , is assumed to be a quadratic function of the
number of Starbucks café shops present in the local market, i.e.: 8
 2
 0 = 0 + 1 ℎ + 2 ℎ +∑ ∈ 3, ( × ) (6)
where Trend is a time trend variable based on year-month combinations; is a zero-one
local area zip code dummy variable; is the set of zip codes, i.e., distinct local areas, in our data;
and 3, is a zip code-specific parameter that captures the composite impact of local market-
specific trends on the mean utility obtained from the outside option. Therefore, the last term in the
above equation controls for local market-specific time-varying factors, which may include the
market presence of competing non-Starbucks café shops, that influence consumers’ mean utility
obtained from the outside option for the relevant market. Parameters 1 and 2 together capture
the potentially non-linear marginal impact on utility obtained from the outside option due to the

8
 In studying the effect of ownership structure and market geography on prices of fast-food chains (McDonald’s vs.
Burger King), Thomadsen (2005) model the mean utility for the outside good as a linear function of consumers
demographic profile, including age, gender and race, etc. In fact, we also estimated the demand with the outside good
mean utility as a linear function of Nshop.

 17
number of Starbucks café shops present in the local market. Last, 0 and 0 are normalized to
be zero.
 Our characterization of the “business-stealing” effect is the situation in which the
increasing market presence of Starbucks café shops more easily facilitates consumers stopping by
one café to grab a cup of coffee and thus reduce the purchases of packaged ground coffee from
grocery stores. From the perspective of our empirical demand model that focusses on packaged
ground coffee in grocery stores being the “inside” goods, with coffee consumption at café shops
being part of the “outside” option, the “business-stealing” effect corresponds to the mean utility
obtained from the “outside” option being positively influenced by the increasing market presence
of café shops. Specifically, in our empirical model the “business-stealing” effect implies
 0 
 > 0 from equation (6), which is determined by parameters 1 and 2 , while in our
 ℎ 

theoretical framework it implies that ′ ( ) < 0 from equation (1).
 In summary, our structural empirical demand model is specified to separately identify the
“demand-increasing” and “business-stealing” effects via parameters ( ,1 , ,2 ) and ( 1 , 2 ) ,
respectively.
 Last, the probability that product is chosen, or equivalently the model-predicted market
share of the ℎ product, is the integral over the mass of consumers that select this product:
 + 
 ̂ ( )
 ( , , ; , , , , Γ, Σ) = ∫ + 
 ( ) (7)
 0 +∑ =1

where is the mean utility and is the deviation from the mean utility that allows for
 ̂ and ( ) are population distribution functions for consumer
consumer heterogeneity; ( )
demographics and random taste shocks, respectively, assumed to be independently distributed.9

5.2 Supply
 We assume coffee manufacturers strategically set their prices for the packaged ground
coffee sold in the grocery stores in a non-cooperative way to maximize their profits in a static Nash
equilibrium price-setting game. 10 Suppose multi-product coffee manufacturers compete in

9
 In the actual demand estimation, we use 200 random draws from (∙) for the numerical approximation of (∙).
10
 To simplify the supply-side analysis, we assume retailers do not play a strategic role in setting retail prices of the
coffee products in our analysis, and simply set retail prices just high enough to cover their economic retailing costs
and costs to obtain coffee products from coffee manufacturers.

 18
Bertrand-Nash fashion. Each firm offers a menu of products in market at time , , and
sets prices of these products to maximize its variable profit as follows:
 max = max ∑ ∈ ( − ) (8)
 ∀ ∈ ∀ ∈ 

where is the marginal cost incurred by the firm to provide product in market at time .
The quantity sold of product in equilibrium, , is equal to the market demand of product ,
that is, = = × ( ), where is a measure of the potential size of market m
during period t, ( ) is product ’s predicted market share based on equation (7), and is the
vector of prices for the packaged ground coffee products.

5.3 Estimation and Identification
 Our estimation strategy closely follows the generalized method of moments (GMM)
approach taken by commonly known studies in the Industrial Organization literature, such as BLP
(1995), Nevo (2000a, b), and many others. The key identifying assumption in the estimation lies
in the population moment conditions that are constructed by interacting instrument variables with
the implied structural error term. The estimation algorithm is performed to search the unknow
parameter values such that the observed product shares are equal to the product shares predicted
by the demand model. Detailed discussion on the identification of structural parameter estimates
as well as the procedure of the search algorithm can be found in Nevo (2000b).
Instruments
 The classic econometric problem in discrete-choice demand estimation is the endogeneity
of product prices, as product characteristics that are unobserved by researchers in , are likely
correlated with prices. The groups of fixed effects dummy variables in described above
account for some of the unobserved product characteristics in and therefore substantially help
with mitigating the endogeneity problem.
 As for the exogenous instrument variables, we first use the direct components of marginal
costs (e.g., manufacturer input prices) interacted with brand dummies as in Villas-Boas (2007a,
2007b), Nakamura and Zerom (2010), and Gayle and Lin (2022). We use the composite indicator
prices calculated by the International Coffee Organization (ICO) as a proxy for the raw coffee bean
prices. By interacting the raw coffee bean prices with the brand dummies, we allow the raw bean
prices to influence the production costs differently across brands. Second, we interact the national
average industrial electricity prices with the dummy variables that are generated from the four

 19
different coffee packaging materials in the data.11 These interactions aim to capture the likelihood
that changes in national average electricity prices affect ground coffee packaging costs differently
across different packaging processes. Last, following BLP (1995) estimation procedure, we also
include several instruments for prices based on some observable non-price product attributes of
competing products offered in the market. We include the sum and mean of caffeine content of
these competing products offered in the market.
 To instrument for the random component of product prices associated with consumer-
specific demographics, , we construct three-way interaction instrument variables by using the
above electricity price-packaging material dummies interactions to multiply with county-level
median personal income for the population aged sixteen and above. We are convinced that after
using the various fixed effects to control for a substantial part of the unobserved product attributes
in , then the remaining components in are most likely uncorrelated with county-level
average income.
 To identify the standard deviations of the random coefficients on price, the organic dummy,
and the constant term, we follow Gandhi and Houde (2020) and Gayle and Lin (2022) and
construct product differentiation instruments along three dimensions. 12 These differentiation
instruments are intended to capture the degree of differentiation of a product relative to other
available products offered in the market. The variations in these differentiation measures induce
consumer substitution along these dimensions, and thus identify the standard deviation preference
parameters for the random coefficients. In addition, we also include a standard instrument used by
many empirical industrial organization studies: the total number of coffee products offered in a
market, which identifies the standard deviation preference parameter on the intercept.13
 Another endogenous concern is the measured presence of Starbucks café shops, Nshop.
We instrument for Nshop with several groups of exogenously determined cost-side influencers of
Starbucks café shop market presence, which are expected to be negatively correlated with the
presence of Starbucks café shops. These instruments include the county-level average weekly

11
 Electricity prices for industrial use are sourced from the U.S. Energy Information Administration (EIA) under the
electricity data page. In the coffee data, packaging materials are classified into the four categories: paper bags/boxes,
laminated (foil) bags, plastic canisters, and light metal tins.
12
 For the construction of these differentiation instruments, we refer the readers to Table 12 in Gandhi and Houde
(2020). The predicted prices are obtained from a reduced-form regression of prices on all the non-price product
characteristics as well as the exogenous cost-shifters and fixed effects previously discussed.
13
 See the relevant identifying argument in Sullivan (2020) and Miller and Weinberg (2017).

 20
wage (both for all sectors and private sectors), and state average commercial electricity and natural
gas prices. 14 These instrument variables are considered as the direct components of business
operating costs for Starbucks café shops, i.e., geographic areas incurring higher employee wages
or utility expenses are less likely to incentivize Starbucks café shop entry. Last, according to
Wooldridge (2010), a natural instrumental variable for the interaction between an endogenous
variable and an exogenous variable is to interact the instruments for the endogenous variable and
this exogenous variable. Therefore, the instruments for the interaction term between Starbucks and
Nshop, are simply the product of the above-mentioned cost-side influencers of Nshop and the
Starbucks brand dummy.

6. Empirical Results

6.1 Demand Parameter Estimates

 Three panel of estimates of the demand parameters are reported in Table 3. Panel (1) and
panel (2) present the estimates obtained from a standard logit specification of the demand model
using the ordinary least squares (OLS) estimator and the two-stage least squares (2SLS) estimator,
respectively. The GMM estimates for the random coefficients logit specification are reported in
panel (3).
 The Wu-Hausman statistical test for endogeneity rejects exogeneity of the endogenous
variables discussed above. Without instrumentation the OLS estimate on Nshop has a negative
sign, but a positive sign when using 2SLS and GMM estimators. The OLS estimate on the
interaction between Starbucks dummy and Nshop, though positive, is smaller in magnitude than
the estimate obtained from 2SLS and GMM procedure. This finding therefore supports the need
for instrumentation. Moreover, the Stock and Yogo (2005) statistical test for weak instruments
rejects the null hypothesis that the instruments used in the demand estimation are weak. Following
Gandhi and Houde (2020), we also perform the Independence of Irrelevant Alternatives (IIA)
hypothesis test to further examine the possibility of weak identification problems.15 The IIA joint
test statistics validates the ability of our product differentiation instruments to identify deviations

14
 The lags of these cost-side instruments for up to 3 periods are included as additional instruments. The weekly wage
rates are obtained from U.S. Bureau of Labor Statistics under “Quarterly Census of Employment and Wages” database.
The state average electricity and natural gas prices are both obtained from EIA.
15
 See Gandhi and Houde (2020) for details of how to test for weak identification issues in random coefficients demand.

 21
of the random coefficients from the standard logit preferences. As such, we focus the remainder
of our analysis on the set of demand estimates from the GMM estimator in panel (3) of the table.

 Table 3: Demand Estimates
 Standard Logit Model Random Coefficients Model
 ( = ) ( ≠ )
 (1) OLS (2) 2SLS (3) GMM
 Mean Standard Interaction with
 Mean Coef. Mean Coef.
 Coef. Deviations Income
 ( , , ) ( , , ) ( , , , ) (Σ) (Γ)
 Price ($/oz) -2.474*** -2.076*** -2.7145*** -0.0011 -16.15***
 (0.0148) (0.036) (0.0558) (1.862) (4.146)
 Constanta -4.435*** -6.036*** -7.7504*** 0.7521 1.1856
 (0.0877) (0.107) (0.0481) (1.155) (5.792)
 Organica -0.779*** -0.824*** 0.0129* 0.1935 17.842***
 (0.00854) (0.009) (0.0068) (1.225) (2.668)
 ̂1 )
 Nshop ( -0.0079*** 0.151*** 0.1224***
 (0.00158) (0.0106) (0.0103)
 ̂2 )
 Starbucks×Nshop ( 0.0472*** 0.0694*** 0.0675***
 (0.00169) (0.0039) (0.0046)
 Caffeine 0.369*** 0.374*** 0.3966***
 (0.00231) (0.0024) (0.0023)
 Package Weight 0.0265*** 0.0300*** 0.0245***
 (0.000240) (0.0004) (0.0005)
 Mean Utility of the Outside Goodb
 R-sq 0.468 0.453 Nshop ( ̂1 ) -0.09*** (0.00242)
 Wu-Hausman (F-
 142.549 Nshop2 ( ̂2 ) 0.0045*** (0.00025)
 statistic)
 Stock and Yogo (2005) GMM objective value 7,213.24
 Weak Instrument Test 44.202 IIA joint test on
 7,939.08
 (F-statistic) differentiation IVs (Chi-sq)
 No. of markets 6,780
 No. of observations 402,330
 Notes: ***p
You can also read